diff --git a/osdc/clusters.yaml b/osdc/clusters.yaml
index 39845559..7a3a190c 100644
--- a/osdc/clusters.yaml
+++ b/osdc/clusters.yaml
@@ -53,7 +53,7 @@ defaults:
     pdb_enabled: true
     pdb_min_available: 1
   arc:
-    chart_version: "0.14.1-jeanschmidt.12"  # MUST match for controller + runner charts (minor version mismatch = ARS deletion). Fork chart published from jeanschmidt/actions-runner-controller.
+    chart_version: "0.14.1-jeanschmidt.13"  # MUST match for controller + runner charts (minor version mismatch = ARS deletion). Fork chart published from jeanschmidt/actions-runner-controller. .13 adds WorkflowSchedulerName so the listener applies schedulerName to placeholder pods (numa-scheduler enablement).
     replica_count: 4
     log_level: info
     controller_cpu_request: "1"
@@ -292,6 +292,8 @@ clusters:
       - arc
       - nodepools
       - nodepools-h100        # H100 only — B200 has no capacity reservation in us-west-1
+      - nfd                   # NUMA topology data for p5 nodes
+      - numa-scheduler        # NUMA-aware secondary scheduler
       - arc-runners-h100
       - pypi-cache
       - cache-enforcer
@@ -378,6 +380,8 @@ clusters:
       - nodepools
       - nodepools-b200
       - nodepools-h100
+      - nfd                   # NUMA topology data for p5 nodes
+      - numa-scheduler        # NUMA-aware secondary scheduler
       - arc-runners
       - arc-runners-b200
       - arc-runners-h100
diff --git a/osdc/modules/arc-runners-h100/defs/l-x86iamx-88-900-h100-4.yaml b/osdc/modules/arc-runners-h100/defs/l-x86iamx-88-900-h100-4.yaml
index 7100922e..5e1f6793 100644
--- a/osdc/modules/arc-runners-h100/defs/l-x86iamx-88-900-h100-4.yaml
+++ b/osdc/modules/arc-runners-h100/defs/l-x86iamx-88-900-h100-4.yaml
@@ -7,6 +7,10 @@ runner:
   vcpu: 88
   memory: 900Gi
   gpu: 4
+  # NUMA-aware scheduling: p5.48xlarge is a 2-NUMA-node box (4 H100 per zone).
+  # Without this, the default scheduler may place a 4-GPU pod on a node where
+  # both zones are partially used, causing TopologyAffinityError.
+  scheduler_name: numa-scheduler
   # Fixed-capacity cap: 1 reserved 8-GPU node / 4 GPUs per runner = 2 concurrent runners.
   max_runners:
     default: 2